Gibbs sampling via neural network probability estimation

The authors propose a neural network approach an efficient nonparametric approach, for Markov random field (MRF) modeling to provide a good estimate of Bayesian a posteriori probability. The approach overcomes the difficulties encountered in estimating the parameters of the Gibbs distribution that characterizes the MRFs and the underlying texture. Its successful application to textured image segmentation using the Gibbs sampling technique is shown.<>